Humana, 2024
Building Trust in AI-driven risk predictions
Enhancing transparency and confidence in AI-driven risk predictions, empowering nurses to identify and escalate critical cases swiftly and effectively.
My Role
UX Designer
Led the end to end UX design and was also part of the user research team, conducted Interviews, contextual inquiries, usability testing.
Team
2 UX Designer
2 UX Researchers
2 Developers
1 Product Manager
1 Lead Application Architect
Timeline
8 months
Tools
Figma, Figjam, Zeplin Qualtrics, Adobe Photoshop
Context
Nurses at Humana play a critical role in delivering follow-up care to older adults discharged from hospitals under Medicare. These patients are often at risk of readmission or serious complications if deteriorating conditions go unrecognized. To support nurses in identifying high-risk patients, Humana introduced an AI-powered risk prediction feature in its internal care coordination tool. However, despite the potential of AI to improve patient outcomes, nurses were hesitant to rely on it for making escalation decisions.
Goal: We set out to uncover the reasons for this hesitancy and redesign the tool’s AI interface to foster trust, transparency, and actionable insights.
Platform: Enterprise B2B SaaS for Humana healthcare providers
User Base: RNs, case managers managing post-acute Medicare patients
About the Tool: Nurses use an internal tool to access and manage patient information for older adults on Medicare who require follow-up care after hospital discharge. The tool is designed to support timely, effective care by providing up-to-date health details and care needs.
Follow-up care after hospital discharge
Older Adults
Registered Nurses
The Problem
Nurses lacked trust in AI-generated risk scores due to opaque “black-box” predictions, delaying critical escalations and increasing cognitive load.
The Solution
Redesigned the interface with explainable AI risk scores, centralized high-priority information, and intuitive, action-oriented workflows to foster confidence and speed.
Impact
66%
increase in trust and confidence in AI-driven risk predictions
50%
reduction in time spent escalating critical cases
100%
increase in user satisfaction after the redesign
Design Process in Action
RESEARCH
User Research Overview
The Challenge
Why weren’t nurses trusting — or acting on — the AI risk scores?
Before the Redesign
Before the redesign, nurses had to navigate an inefficient workflow full of confusion and manual effort. Here's what that looked like.
01
Check Dashboard
Information overload creates decision paralysis
😐 Neutral
"There's so much data, I don't know where to start looking"
Cognitive Load: High
02
Assess AI Recommendation
AI felt like a black box - no explanation of the score
😵💫 Confused
"I don't know what's behind the score or if I can trust it"
Trust level: 22%
03
Review Patient Data
.png)
No centralized view of risk context making it time consuming to find the relevant information
😲 Overwhelmed
"The alert says 'elevated risk' but doesn't tell me why"
~7–10 clicks per case
04
Escalate to Doctor
.png)
Communication gaps lead to repeated explanations
😤 Frustrated
"I have to explain the same patient situation three times"
Call Duration: 12 min
04
Document Decision
.png)
Manual documentation creates workflow friction
😩 Exhausted
"By the time I finish documenting, I've forgotten half the details"
Documentation Time: 18 min
Why It Mattered
These observations formed the basis for our redesign — focused on trust, explainability, and guidance. By understanding where nurses struggled most, we could design solutions that truly supported their workflow.
Design Challenge: How Might We...
“How might we simplify workflows to reduce time spent gathering patient information?”
“How might we help nurses understand and trust AI risk scores enough to act confidently?”
UI and Navigation Issues
Turning pain points into user stories
DESIGN VALUE 4: SIMPLICITY & CLARITY
USER STORY 4
"I want to easily navigate in the community when I am looking for my desired content, so I don’t feel lost. "
PAIN POINT 4
“I feel overwhelmed while navigating the Trailblazer community, as the current information architecture is scattered, making the features difficult to find.”
The aim is to streamline the navigation process within the community, making it intuitive and user-friendly so that users can effortlessly find the content they are looking for. Improving the clarity and simplicity of how the information is presented can result in less confusion and boost engagement in the community.
From Insights → Design Principles
01
Consolidation & Clarity
-
Summarize patient info at top
-
Reduce navigation & cognitive load & Clarity
02
Trust & Explainability
-
Show risk factors transparently
-
Build trust in AI & guide action
Achievements
led to a 66% increase in trust
Nurses feel confident using the risk score, knowing exactly how it was determined and what to do next.
Average time to escalate a critical case dropped by over 50%
With clear, actionable insights, high-risk cases are escalated more promptly and consistently.
Improved Satisfaction by 100%
Nurses completed all critical tasks successfully and felt less cognitive load, indicating a higher likelihood of timely interventions
Lesson Learned
Stakeholder Engagement is Essential
Early and ongoing conversations with nurse managers, compliance, and IT stakeholders ensured that the solution was not only user-friendly but also aligned with organizational goals and regulatory requirements.
Cross-Functional Collaboration Drives Success
Working closely with engineers, the PMs, and the UX researcher in daily standups and Agile sprints accelerated problem-solving & helped surface edge cases early. This collaboration ensured that technical constraints were addressed proactively.
User Feedback is Invaluable
Regular usability testing and feedback sessions with nurses highlighted pain points that weren’t obvious at the outset. Iterating quickly based on their input resulted in a more intuitive and trusted tool.
Like what you see?
Let's get in touch
© 2025 Created with love & lots of coffee by Aneesha Chinni 💗☕